We offer MATLAB code to take a normal, 2-D video and automatically compute depth at every frame. We also offer a database to help users learn the association of 2-D to depth.
Note By installing, copying, or otherwise using this software, you agree to be bound by the terms of its license. Read the license.
Installation Instructions for MSR-V3D database
Download the Matlab code by clicking the download button above.
The following are the training database files. These are large files. You can download them by clicking the links below.
- Indoor 1: ftp://ftp.research.microsoft.com/downloads/29d28301-1079-4435-9810-74709376bce1/StereoRGBD_Indoor_1Ap1.zip
- Indoor 2: ftp://ftp.research.microsoft.com/downloads/29d28301-1079-4435-9810-74709376bce1/StereoRGBD_Indoor_1Ap2.zip
- Indoor 3: ftp://ftp.research.microsoft.com/downloads/29d28301-1079-4435-9810-74709376bce1/StereoRGBD_Indoor_1Bp1.zip
- Indoor 4: ftp://ftp.research.microsoft.com/downloads/29d28301-1079-4435-9810-74709376bce1/StereoRGBD_Indoor_1Bp2C.zip
- Indoor 5: ftp://ftp.research.microsoft.com/downloads/29d28301-1079-4435-9810-74709376bce1/StereoRGBD_Indoor_234.zip
- Outdoor: ftp://ftp.research.microsoft.com/downloads/29d28301-1079-4435-9810-74709376bce1/StereoRGBD_Outdoor.zip
If you use our database for your research work, please cite our paper:
K. Karsch, C. Liu, and S.B. Kang, "Depth extraction from video using non-parametric sampling," European Conference on Computer Vision, Oct. 2012.
Information on MSR-V3D database
The Microsoft Research Stereo Video + Depth (MSR-V3D) database contains stereo video clips for a variety of scenes and viewpoints (116 indoor, 61 outdoor). The dataset primarily contains videos recorded from a static viewpoint with moving objects in the scene (people, cars, etc.). There are also 100 indoor frames in addition to the 116 indoor clips within the database (stereo RGB, depth for the left frame).
We captured the MSR-V3D dataset with two side-by-side Microsoft Kinects. For each indoor clip, the left stereo view also contains view-aligned depth from the Kinect. Due to IR interference, depth for the right view was not captured indoors, and depth was totally disregarded outdoors due to limitations of the Kinect.
One use of this dataset is for training and evaluating our Depth Transfer algorithm (code above), but we hope that it may be useful in other scenarios as well. For additional details concerning the dataset (including MATLAB code for reading/viewing the data), please see the README file within the zip files.